How to appear in ChatGPT

ChatGPT names a brand by two different paths, from memory and by live search, and each one demands different work. The mechanism, what to do for each, what you can verify on your site today, and how long each path takes to move.

Start with this, because it changes everything else: ChatGPT names a brand by two different paths, and most of what is written on the subject treats the two as if they were one.

The two paths

From memory. The model answers with what stayed from training. It does not go to the internet, it cites no link, and it says names of brands it learned to associate with a type of need. Here whoever is a recognised entity across many independent sources wins. It is slow to change: it depends on training cycles that happen when OpenAI decides.

By live search. The question triggers a search, the model reads pages at that moment and answers citing them. Here whoever has findable content that answers the question wins, and whoever is present on the pages the engine chooses to read. It is fast to change: weeks, not months.

The practical consequence is simple. If you need a result this quarter, work the second path. The first is built in parallel and pays later, for longer. We wrote about the difference in detail in knowledge vs augmented.

What to do, in order

01 · Open the door to both robots, which are not the same

OpenAI uses distinct agents with distinct purposes. GPTBot collects content that may enter future training. OAI-SearchBot fetches pages at the moment somebody asks a question with search active. Blocking the first is a defensible business decision. Blocking the second is choosing not to exist in today's answers.

Do: in robots.txt, allow at least OAI-SearchBot. Decide on GPTBot with a cool head: if you live on being found, allowing it tends to pay off.

02 · Be an entity, not just a site

The model does not store pages, it stores entities and what is said about them. A brand that exists only on its own domain is an assertion without confirmation. A brand with records in public databases and consistent profiles is a fact verifiable in several independent sources, and that is how it comes to be named from memory.

Do: declare sameAs in the organisation schema pointing at profiles that actually exist, and see to it that you have them: a Wikidata record, a company LinkedIn, and the directories that carry weight in your sector. Name, address and phone the same in all of them, to the character.

03 · Publish what only you can publish

An engine cites whoever adds something. Text that repeats what already exists on twenty sites gives no reason at all to be chosen. Original numbers, a declared methodology, cases with a verifiable result and honest comparisons, including with whoever is better than you at something, are what stands out among interchangeable content.

Do: pick a question your buyer actually asks and answer it better than any other page in your language. One a month is worth more than four generic ones.

04 · Be in the sources the engine reads when it searches

In live search mode ChatGPT does not make things up: it reads pages and cites them. Often those pages are not yours, they are lists, comparisons, forums and press. Appearing well on those sites is frequently faster than trying to get your own page cited.

Do: find out which sources are cited in the questions of your category and work on presence in those, one by one. It is relationship work, not code.

05 · Measure, because the answer changes and does not warn you

The same question gives different answers in different weeks, and nobody sends you an email when you stop being cited. Without repeated measurement you cannot tell whether the work paid off, and the discussion runs on impressions.

Do: fix a set of real buyer questions and run them on a cadence, keeping the history. It is what we do, and it is what lets you say whether an action moved the needle.

The checks, in two commands

So that this does not stay theory, here is what is verified with two commands. Swap the domain for yours.

The two OpenAI agents have a way in

$ curl -s https://example.com/robots.txt | grep -iA1 'GPTBot\|OAI-SearchBot'

User-Agent: GPTBot
Allow: /
--
User-Agent: OAI-SearchBot
Allow: /

Both allowed, each with its own block. If you only have a generic rule with Disallow, the two are left out and the rest of the work never gets to count.

The brand declares itself as an entity

$ curl -s https://example.com/ | grep -o '"sameAs":\[[^]]*\]'

https://www.linkedin.com/company/example
https://www.crunchbase.com/organization/example
https://www.wikidata.org/wiki/Q000000

Independent sources, and among them Wikidata, which is the one that weighs most because it is read by practically every system that builds entity graphs. A warning from someone who has been through it: the Wikidata item is only created after public references exist (press, studies, official records). An item without references is deleted for lack of notability, and a sameAs pointing at a deleted item is worse than none. If this command returns nothing, the brand exists only by its own assertion.

The full check on whether the site is legible is in six commands to find out whether the AI can read your site.

What does not work

There is no submission form for ChatGPT, there is no brand panel, and you do not pay to be recommended in the organic answer. If somebody sells you registration, they are selling something that does not exist.

Filling the site with generated pages does not work either, and today it is counterproductive: interchangeable content is exactly what an engine has no reason to cite. And declaring schema that does not match what is on the page gets caught, and costs more than having no schema at all.

Frequently asked questions

Read next

• Your B2B SaaS does not appear in ChatGPT, the same problem seen from one sector.
• How to appear in Google AI Mode, the equivalent on the Google side.
• What entity authority is, the second step in detail.